Skip to content
AI-inzichten & Trends

Building your own legal AI: what you make and what you can reuse

Beau Jonkhout

Dutch law firm NautaDutilh is building its own AI applications with Bluefield. A good move. What sits underneath such an application, which part is the same for every firm, and how you will build on the Prudai Developer API, starting with free Dutch legal sources.

Close-up van drie naast elkaar staande ordners in beige, geel en groen met ronde vingergaten in de rug

On 16 October 2026, the Dutch law firm NautaDutilh announced that it will develop its own AI applications under the name Bluefield. Lawyers and software engineers build together, on the firm's own data and alongside tools such as Legora that it already uses. That is a good move. A firm that builds its own applications shows that AI is becoming part of how legal work gets done. Anyone heading the same way does not have to build everything themselves. Sources, citation checks, processing, security and the homework for the AI Act and the GDPR are largely the same for every firm. That foundation can be reused. That is why we are opening the Prudai Developer API, starting with a free sources API over Dutch open legal sources.

In short

  • NautaDutilh launched Bluefield: its own platform on which lawyers and software engineers build AI applications together, for the practice and for clients.
  • Elgar Weijtmans (AI & Innovation Partner) and Niels de Jong (AI & Innovation Lead) start on 1 November and lead Bluefield together with three partners.
  • Anyone who develops an AI system and puts it into service under their own name is a provider under the AI Act, even if the system only runs internally.
  • The Prudai Developer API starts with free Dutch open legal sources: search with verified passages and a citation check that also checks other models. LEO as a model follows. Anyone with an account creates their own keys.

What did NautaDutilh announce?

Bluefield is the firm's own platform for developing technology for legal practice. Lawyers and software developers work together on solutions that come from concrete questions in daily practice. The platform works with the firm's own data, within its own requirements for information security and confidentiality. Bluefield complements the AI applications the firm already uses, such as Legora, and does not replace them.

Elgar Weijtmans joins from HVG Law on 1 November as AI & Innovation Partner and leads the AI & Innovation team. Niels de Jong becomes AI & Innovation Lead, responsible for software development. Together with partners Jinne van Belle, Sybren de Beurs and Marc Orval, they steer Bluefield. Lawyers, notaries and tax advisers remain responsible for the services and for reviewing the results, Advocatie reports.

Why is this a good move?

"The greatest opportunities for AI in the legal sector lie close to legal work itself," Weijtmans told Advocatie (our translation). We agree. A general AI tool does not know how your team builds a file, which clauses your firm uses as standard, or what a regular client expects from a memo. That knowledge sits in the firm. A team that brings lawyers and engineers together around real questions can build exactly there.

In the Dutch financial daily FD, legal consultant Pim Betist expects more Dutch firms to build their own AI platforms. AI thus becomes a fixed part of the work, not just something a firm buys.

What sits underneath your own AI application?

The application a lawyer sees rests on layers they do not see. The Innovative Lawyer listed a number of them on 20 October.

Sources. Legislation and case law have to be current. A statute changes, a judgment is overturned on appeal. Whoever opens up sources themselves also maintains the connections and the licences.

Citations. A language model can name an ECLI that does not exist or belongs to a different judgment. A check that verifies whether a cited judgment was actually retrieved has to be built and maintained by someone. Why that matters is explained in our blog on disciplinary law and AI citations.

Processing and contracts. Where does the model run, does the provider train on your input, and how do you switch if you need to? Those are agreements with the model provider, and they have to match what you promise your clients.

AI Act. Anyone who develops an AI system and puts it into service under their own name is a provider, also for own use. If an application talks directly to clients, transparency obligations apply. According to The Innovative Lawyer, an internal tool for research or drafting generally falls outside the high-risk category, but you assess that per application. More in our blog on the AI Act and high-risk systems.

GDPR and confidentiality. Personal data in a case file was collected to assist a client. Using it to develop or test a tool means using it for a different purpose, and you check whether that is compatible. On top of that, the duty of confidentiality applies, also to anyone involved in the work as an engineer.

What can you reuse?

A large part of those layers is the same for every firm. It does not set you apart, but it does cost engineering time. What remains is yours.

LayerWhat the foundation providesWhat stays with your firm
SourcesMore than 100 sources in LEO, including legislation and case law, kept up to date; some are queried live at the sourceYour own model documents, templates and knowledge
CitationsAnswers with source and location, and a check on ECLI numbersThe substantive test: does the judgment support your argument?
ProcessingProcessing and storage of customer data take place exclusively within the EU; the model providers contractually do not train on your inputWhich data you send to your application
SecurityPrudai is certified to ISO/IEC 27001:2022 and NEN 7510-1:2024The security of your own application
AI Act and GDPRDocumentation of the foundation: how the model works, which sources it uses, where it processesYour role and your risk assessment per application

How it works at Prudai: building on the foundation

The Prudai Developer API comes in two steps. Anyone who creates an account creates their own keys.

First: free Dutch sources. The first step is a free sources API over Dutch open legal sources.

  • Search with verified passages. Every search result returns the exact passage, its location and the literal quote.
  • Fetch the full text where the source's licence allows it.
  • A citation check for any model. Send in ECLIs, statutory articles or parliamentary papers, with or without a quote. You get back whether the reference exists, what it points to and whether the quote matches. It also works on the output of other AI models.
  • Free at launch, with fair use as the limit.
  • Also via MCP, so AI assistants and agents can consult the sources directly.

Then: LEO as a model. The chat API follows the OpenAI message format. You choose LEO as the model, or a specialisation such as employment law. Every answer contains structured citations and states what it cost. You can enable your organisation's knowledge bases per key, so LEO also searches your own firm knowledge.

Open-source SDK. An open-source SDK for Python and TypeScript is coming, so your developers can start with a few lines of code.

Your engineers then build the application that fits your firm: an intake that asks your own questions, a draft in your own structure, a connection to your document system or an environment for your clients. The legal foundation underneath comes from us. LEO researches and substantiates; the lawyer assesses and remains responsible.

For the specialist

  • Provider. Article 3, point 3 of the AI Act: whoever develops an AI system (or has one developed) and places it on the market or puts it into service under their own name. According to point 11, "putting into service" also covers supply for own use. The deployer is defined in point 4.
  • Transparency. Article 50(1): systems intended to interact directly with natural persons must be designed so that those persons know they are dealing with AI.
  • Compatibility. Article 6(4) GDPR lists, among other things, the link between the purposes, the context of collection, the nature of the data, the consequences for data subjects and appropriate safeguards.
  • Confidentiality. Article 11a of the Dutch Advocates Act extends the duty of confidentiality to staff, personnel and other persons involved in the practice of the profession.
  • The API, technically. Sources: search with the verified passage per result, fetch where the licence allows it, and a citation check for ECLIs, statutory articles, CELEX numbers and parliamentary papers. Chat: OpenAI-compatible chat completions; the model is called leo or, for example, leo/arbeidsrecht.

Frequently asked questions

Is the Prudai Developer API available yet?

Not yet. The first step will be the free sources API over Dutch open legal sources, followed by LEO as a model. Anyone who creates an account creates their own keys. Want to be among the first? Request early access.

What does it cost?

Search, fetch and the citation check are free at launch, with fair use as the limit. For the chat API with LEO you pay per request; every answer states what it cost.

Does a firm still need to arrange anything for the AI Act itself?

Yes. For your own application, you determine which role you have and which risk category the application falls into. We provide the information about the foundation that helps with that assessment.

Does an in-house platform replace the AI tools a firm already uses?

Not at NautaDutilh: Bluefield complements tools such as Legora. You decide per application what you buy and what you build.

Does the lawyer remain responsible?

Yes. LEO researches and substantiates, with sources you can check. The lawyer assesses the outcome and remains responsible for the advice. NautaDutilh says the same about Bluefield.

Sources

Building on free Dutch legal sources, with passages and citations you can check? Request early access to the Prudai Developer API.

AI ActMCPAI in organizationsPrudai

Beau Jonkhout

Technical Director

Beau is co-founder and technical director of Prudai. He is the driving force behind the technical architecture of the Prudai platform. He leads the development of the multi-agent frameworks, manages the developers, and is responsible for the integration quality, security, and privacy by design of all solutions.